-
Notifications
You must be signed in to change notification settings - Fork 74.7k
Wrong metadata of tensorflow-cpu-aws in TensorFlow 2.16.1 #63585
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Comments
Any updates? This issue makes it impossible to install tensorflow-cpu (any version) on Python 3.12 when using Poetry. Or is there a workaround? For example, when using this
And then running
|
This also affects other Python versions, not just 3.12. I'm on 3.10. How am I supposed to depend on this version of TensorFlow when using poetry? This seems to be broken?!? |
This also completely prevents creating a lock file with [project]
requires-python = ">=3.11,<3.12"
name = "foo"
version = "0.0.1"
dependencies = [
"tensorflow>=2.16.1",
] then
verbose log
|
Installing tf-nightly-cpu version 2.17.0.dev20240412 with Poetry works, so this is fixed in master (intentionally?). If patch release 2.16.2 is going to be made, the fix of this issue should be included in it. I tried to identify the fix, but could not understand the publish/build process. Edit: Poetry shows only one dependency for tf-nightly-cpu:
Whereas for tensorflow-cpu v2.15.1 Poetry showed about 20 dependencies. 🤔 |
TL;DR: Poetry encounters dependency resolution issues with I created a sample project using Poetry to investigate this issue. I found that Poetry successfully resolves dependencies when using Specifically, the following configuration works:
While this configuration does not:
As others have noted, the problem arises when resolving the dependency on
Upon inspecting the metadata of both
It's unclear why depending on I've applied this fix to all
@randolf-scholz While I haven't tested it, it seems that this approach should also work with pdm, based on their documentation. I hope this temporary solution helps until the release of TF 2.16.2 (expected in the next month or two). Please let me know if anyone encounter any further issues. tensorflow
tensorflow-cpu
|
@nitins17, thanks for addressing this. Unfortunately, this would not work for us as we are not using poetry and the patched wheel file is for 3.11 only. Just wondering when can we expect the release of 2.16.2? This is blocking us from supporting Python 3.12 as we can't use 2.16.* in our dependencies. |
@tachikoma-li Patched versions of all supported wheels should be present in
E.g: If you want to install the patched Python 3.12 wheel of TF 2.16 for Linux x86 then run the following:
Sadly, I know this isn't the cleanest of solutions but hope it helps until TF 2.16.2 comes out.
Branch cut was done this week. I'm guessing here but I think rc0 should come out in the next week or so and then 3-4 weeks more for the final release provided everything goes smoothly. |
Hey @nitins17, I have the same problem with Thanks! |
TF 2.16.2 has been released! Closing this now as it should be fixed in 2.16.2. Please feel free to reopen if anyone is still running into this issue after upgrading to TF 2.16.2
@miljkovicm15 For Apple Silicon wheels, please switch your workflow to install |
Issue type
Build/Install
Have you reproduced the bug with TensorFlow Nightly?
No
TensorFlow version
2.16.1
OS platform and distribution
Linux x86_64
Python version
3.11
Current behavior?
The metadata of TensoFlow 2.16.1 (for Linux x86_64) has a dependency of
tensorflow-cpu-aws
:However, the latest version of
tensorflow-cpu-aws
is 2.15.0. This would cause trouble when resolving the dependencies with the package metadata.If we can't ensure the release of
tensorflow-cpu-aws
is synchronized withtensorflow
, can we relax the dependency requirement in the metadata?The text was updated successfully, but these errors were encountered: